REGR: to_datetime with non-ISO format, float, and nan fails on main (#50238)

check for nan/nat in floats

Co-authored-by: MarcoGorelli <>

Author: MarcoGorelli

doc/source/whatsnew/v1.5.3.rst

@@ -19,6 +19,7 @@ Fixed regressions
 - Enforced reversion of ``color`` as an alias for ``c`` and ``size`` as an alias for ``s`` in function :meth:`DataFrame.plot.scatter` (:issue:`49732`)
 - Fixed regression in :meth:`SeriesGroupBy.apply` setting a ``name`` attribute on the result if the result was a :class:`DataFrame` (:issue:`49907`)
 - Fixed performance regression in setting with the :meth:`~DataFrame.at` indexer (:issue:`49771`)
+- Fixed regression in :func:`to_datetime` raising ``ValueError`` when parsing array of ``float`` containing ``np.nan`` (:issue:`50237`)
 -
 
 .. ---------------------------------------------------------------------------

pandas/_libs/tslibs/strptime.pyx

@@ -42,7 +42,11 @@ from pandas._libs.tslibs.np_datetime cimport (
     pydatetime_to_dt64,
 )
 from pandas._libs.tslibs.timestamps cimport _Timestamp
-from pandas._libs.util cimport is_datetime64_object
+from pandas._libs.util cimport (
+    is_datetime64_object,
+    is_float_object,
+    is_integer_object,
+)
 
 cnp.import_array()
 
@@ -185,6 +189,12 @@ def array_strptime(
         elif is_datetime64_object(val):
             iresult[i] = get_datetime64_nanos(val, NPY_FR_ns)
             continue
+        elif (
+                (is_integer_object(val) or is_float_object(val))
+                and (val != val or val == NPY_NAT)
+        ):
+            iresult[i] = NPY_NAT
+            continue
         else:
             val = str(val)
 

pandas/tests/tools/test_to_datetime.py

@@ -135,6 +135,17 @@ def test_to_datetime_format_YYYYMMDD_with_nat(self, cache):
         result = to_datetime(ser2, format="%Y%m%d", cache=cache)
         tm.assert_series_equal(result, expected)
 
+    def test_to_datetime_format_YYYYMM_with_nat(self, cache):
+        # https://github.com/pandas-dev/pandas/issues/50237
+        ser = Series([198012, 198012] + [198101] * 5)
+        expected = Series(
+            [Timestamp("19801201"), Timestamp("19801201")] + [Timestamp("19810101")] * 5
+        )
+        expected[2] = np.nan
+        ser[2] = np.nan
+        result = to_datetime(ser, format="%Y%m", cache=cache)
+        tm.assert_series_equal(result, expected)
+
     def test_to_datetime_format_YYYYMMDD_ignore(self, cache):
         # coercion
         # GH 7930